Methods for Stabilization Observation and Identification of Unmanned Vessel Near-shore Navigation Mapping and Shoreline Constraint

Abstract

This invention discloses a method for stable observation identification and shoreline constraint in near-shore navigation and mapping of unmanned surface vessels (USVs), belonging to the field of autonomous navigation and mapping technology for USVs. It is used for near-shore navigation and mapping of USVs, including constructing the USV's state vector and building geometric features near the water surface; obtaining semantic labels for laser points; constructing a near-shore observation disturbance index based on disturbance indicators, calculating semantic and geometric consistency scores, and calculating temporal stability scores; constructing stable observation confidence scores and classifying observations based on these scores; constructing an objective function and setting conservative mapping strategy trigger conditions; and constructing the basic data for subsequent navigation and mapping. This invention, through the deep fusion of stable observation identification and shoreline constraint, effectively suppresses the influence of unstable water surface observations and attitude disturbances, significantly improving the stability, accuracy, and robustness of USV navigation and mapping in complex near-shore environments.

Description

Technical Field

[0001] This invention discloses a method for stable observation and identification and shoreline constraint of unmanned surface vessel (USV) near-shore navigation and mapping, belonging to the field of unmanned surface vessel autonomous navigation and mapping technology. Background Technology

[0002] Unmanned surface vessels (USVs) require stable perception of shorelines, water boundaries, and static structures on land for tasks such as harbor patrols, nearshore mapping, reservoir inspections, inland waterway patrols, maritime supervision, and environmental monitoring. Conventional USV positioning and mapping typically rely on sensors such as GNSS, IMU, lidar, and cameras. However, nearshore waters differ significantly from structured land environments: the water surface lacks stable geometry, and lidar point clouds are prone to specular reflection, wave noise, foam echoes, and wispy discrete points; nearshore structures such as shorelines, wharves, and embankments suffer from discontinuities, weak textures, low profiles, and localized occlusion; and hull roll, pitch, and wave disturbances can cause point cloud distortion, sensor instability, and errors in multi-frame data correlation.

[0003] Existing LiDAR-IMU SLAM methods typically rely on stable edge and planar features, achieving good results in structured terrestrial environments. However, they are prone to the following problems in nearshore waters: unstable points on the water surface enter the static map, forming false walls, trailing lines, ghosting, and floating noise layers; weakly structured shoreline areas lack effective features, leading to pose estimation degradation, map drift, or keyframe error expansion; height threshold filtering alone is insufficient to adapt to changes in hull attitude, water level, and shoreline height; image semantic segmentation alone is easily affected by factors such as water surface reflection, exposure changes, shadows, and blurring of distant shorelines; and ordinary SLAM output is usually a cumulative point cloud map, lacking differentiation between shoreline structure, stable observation confidence, and unstable water surface observations, making it difficult to directly provide a reliable foundation for subsequent unmanned surface vessel navigation map construction.

[0004] Therefore, it is necessary to propose a stable observation-constrained localization and mapping method for unmanned surface vessels (USVs) in nearshore waters. This method unifies the modeling of laser point cloud geometry, image-based shoreline semantic priors, IMU attitude perturbations, and multi-frame temporal consistency to form a stable observation confidence level. This confidence level is then used to drive point cloud matching weights, map update weights, shoreline soft constraints, and adaptive robust mapping strategies. This reduces the impact of water surface reflection, wave trails, and low-confidence observations on static structure maps, thereby improving the stability and cleanliness of USV localization and mapping in nearshore weak-structure environments. Summary of the Invention

[0005] The purpose of this invention is to provide a method for stable observation and identification and shoreline constraint for unmanned surface vessels (USVs) in near-shore navigation and mapping, in order to solve the problems of unstable navigation and mapping and low accuracy caused by dynamic interference on the water surface, attitude disturbance and lack of shoreline constraint in the complex near-shore environment of the prior art.

[0006] Methods for stable observation and identification of unmanned surface vessels (USVs) in near-shore navigation mapping and shoreline constraint include: S1. Construct the unmanned vessel state vector and project the laser points of the shipborne lidar onto the image plane; obtain the neighborhood geometric features of the laser points, construct the water surface proximity geometric features based on the neighborhood geometric features; obtain the semantic labels of the laser points; S2. Construct a nearshore observation disturbance index based on disturbance indicators, calculate semantic and geometric consistency scores based on water surface proximity geometric features, semantic labels, and point cloud geometric static structure scores; divide the nearshore space into map units, calculate temporal stability scores based on the hit status of each map unit; construct stable observation confidence scores, and classify observations based on stable observation confidence scores; S3. Based on the confidence level of stable observations, calculate the map update weights, adjust the point cloud matching information matrix, and perform dual suppression on the map update weights; construct an overall objective function that includes inertial measurement unit pre-integration, laser point cloud registration, shoreline soft constraints, and global navigation satellite system residuals for joint optimization; trigger a conservative mapping strategy based on environmental observability, disturbance index, and stable observation ratio, and implement threshold protection for pose translation and heading jump; S4. Construct a clean structure map based on the shoreline structure, construct a shoreline structure layer based on shoreline geometric elements, construct water surface suppression results based on unstable water surface observations, and construct a stable observation confidence layer based on the confidence of each observation unit; by integrating the clean structure map, shoreline structure layer, water surface suppression results, and stable observation confidence layer, the basic data for subsequent navigation mapping is formed.

[0007] S1 includes S1.1, an unmanned surface vessel equipped with a shipborne lidar, inertial measurement unit, camera, and global navigation satellite system; Map coordinate system is The ship's coordinate system is The lidar coordinate system is The camera coordinate system is The coordinate system of the inertial measurement unit is ; Build The state vector of the unmanned vessel at time t is : ; In the formula, This indicates the position of the unmanned vessel in the map coordinate system. For speed, For attitude quaternions, , For accelerometer bias, For gyroscope bias It is the transpose symbol. For discrete time points; Projecting the laser points of the lidar onto the image plane: ; In the formula, Let be the pixel position of the laser point on the image plane. For the camera intrinsic parameter matrix, For perspective projection functions, This is a transformation from the camera coordinate system to the ship's coordinate system. This is a transformation from the lidar coordinate system to the ship's coordinate system. The laser point in the lidar coordinate system. For laser points, Index for laser points.

[0008] S1 includes, S1.2, and the pair. Extracting the neighborhood Calculate the neighborhood mean Covariance Matrix ,right Perform eigenvalue decomposition and define neighborhood geometric features, including linearity. Flatness Dispersion Local roughness High-altitude characteristics and density characteristics Then build Geometric features near the water surface : ; In the formula, It is an exponential function. For absolute values, For high attenuation scale parameters; S1 includes, S1.3, obtaining Semantic tags, including water surface semantic priors Semantic priors of shoreline proximity and static semantic prior .

[0009] S2 includes S2.1, and disturbance indices including the ship attitude disturbance normalization index. Unstable points on the water surface Degree of degradation of weak structures Semantic uncertainty Constructing a nearshore observation disturbance index : ; ; In the formula, , , , These are the weighting coefficients for the four disturbance indicators. This is a truncation function; S2 includes S2.2, which calculates the semantic and geometric consistency score. : ; ; In the formula, , , , , , The weighting coefficients obtained from the calibration, , , , , These are the weighting coefficients in the geometric static structure score. The score is based on the static geometric structure of the point cloud. For the Sigmoid function, For indicator functions, This is a preset height threshold.

[0010] S2 includes S2.3, which divides the nearshore space into map units. For each map unit... Update the static hit status, shoreline hit status, and water surface hit status: ; ; ; In the formula, for Static hit state at all times for Current shoreline hit status. for Constant water surface hit status, As the consistency score threshold, The semantic probability threshold for the shoreline. The linearity threshold, Threshold for water surface geometric features The semantic probability threshold for the water surface. The attenuation coefficient is... For logical AND operator, For logical OR operator; Calculate the timing stability score : ; In the formula, , , This represents the weighting coefficient in the time series stability score.

[0011] S2 includes S2.4, calculating the confidence level of stable observations. : ; In the formula, , , , , , To stabilize the weighting coefficients in the observation confidence score, Scoring for geometric stability, Score for semantic and geometric consistency. The time series stability score is given. The shoreline structure is a priori. The score is for unstable water surface observations. The degree of unreliability caused by attitude disturbances; Set a high confidence threshold and low confidence threshold ; when , The observation was determined to be highly reliable and stable, and entered into the clean structure map to participate in pose optimization. when , The observation is identified as pending confirmation and enters a short-term buffer while awaiting timing confirmation. when , Observations deemed low-confidence will not be included in the static map, or will only be included in the debugging or suppression layer.

[0012] S3 includes, S3.1, based on Calculate map update weights, adjust the point cloud matching information matrix, and apply dual suppression to map update weights: ; ; ; In the formula, for Map update weight, To update the weights for the minimum map size, For maximum map update weight, for Information matrix used in laser point cloud matching To prevent extremely small positive numbers from being zero, for The nominal information matrix, The weights are updated for the final map after double suppression. For point The confidence level for classifying an observation as unstable on the water surface; S3 includes S3.2, constructing the overall objective function: ; ; ; ; In the formula, The sequence of states to be optimized. This is the optimal state sequence. , and To replace the variable, For the pre-integrated residual of the inertial measurement unit, This is the information matrix corresponding to the pre-integrated residual of the inertial measurement unit. For a priori residuals, The information matrix corresponding to the prior residuals, To register residuals for laser point clouds, To register the information matrix corresponding to the residual of the laser point cloud. As the soft constraint weight of the shoreline, For shoreline soft-constraint residuals, For Global Navigation Satellite System residuals, Let be the information matrix corresponding to the residuals of the global navigation satellite system, and be the robust kernel function. For the first The set of indices of high-confidence points in the laser point cloud used for registration at any given time. For the first The set of indexes for the soft-constraint residuals of the shoreline at any given time. For the first A collection of indexes of observations from the global navigation satellite system at any given time. for The index; The optimized pose of the current frame is obtained by solving the overall objective function. .

[0013] S3 includes S3.3, constructing candidate shoreline points. : ; In the formula, To stabilize the observation confidence threshold, The semantic probability threshold for the shoreline. This is the threshold for shoreline hit status. The linearity threshold, For laser points The index of the map unit to which it belongs; right After fitting, a set of shoreline structures is obtained. : ; In the formula, For the first a section of shoreline, This represents the total number of shoreline segments; For the current shoreline point and map shoreline section The distance is: ; In the formula, For the first A section of the map's shoreline. , For shoreline segment indexing, , and for The two endpoints, for Coordinates in map coordinate system M and for and Coordinates in map coordinate system M This refers to the cross product operation of three-dimensional vectors. shoreline soft constraint residuals for: ; In the formula, for arrive distance, This is the normalized scale parameter for the soft constraint distance of the shoreline; shoreline soft constraint weight for: ; In the formula, For shoreline quality, For the current environmental observability, This is the scaling factor.

[0014] S3 includes S3.4, which defines the environmental observability score. : ; In the formula, , , , , , These are the weighting coefficients in the environmental observability score. To normalize the number of edge features, To normalize the number of planar features, To stabilize the observation ratio, For the current frame's shoreline quality, For attitude perturbation. This represents the proportion of unstable points on the water surface. Define conservative mapping strategies, including low-confidence observation weight decay, keyframe insertion interval adjustment, low-confidence area map expansion limit, and unreliable shoreline soft constraint weight reduction; Set environmental observability thresholds Disturbance index threshold and stable observation ratio threshold ,when or or At that time, a conservative mapping strategy was adopted; Define pose translation jump protection: ; In the formula, This represents the change in pose translation between two adjacent frames. To determine the maximum reasonable translation speed for the unmanned vessel, The time interval between two adjacent frames. To protect the tolerance margin during translation; Define pose and heading jump protection: ; In the formula, This represents the change in heading angle between two adjacent frames. The maximum reasonable heading angular velocity of the unmanned vessel. For heading protection tolerance margin; Define pose translation jump protection threshold and pose and heading change protection threshold : ; ; In the formula, To determine the maximum reasonable translation speed for the unmanned vessel, The maximum reasonable heading angular velocity of the unmanned vessel. The time interval between two adjacent frames. To protect the tolerance margin during translation, For heading protection tolerance margin; when or If the conditions are met or ,accept , The update introduces a jump decay factor. And prevent keyframes from being inserted into the current frame: ; when or If the conditions are met or ,reject It then reverts to the pose predicted by the inertial measurement unit.

[0015] S4 includes shore structures such as wharves, quay walls, railings, and embankments; The geometric elements of the shoreline include shoreline points, shoreline segments, shoreline curves, and shoreline framework; Unstable water surface observations include candidate observations of water surface reflection that have been downweighted, delayed in confirmation, or not included in the clean structure map, candidate observations of wave trails, candidate observations of foam echoes, and low-confidence water surface observations. The confidence level of each observation unit includes the observation confidence level of each point cloud point, raster, voxel, or local sub-map unit; Build the foundational data for subsequent navigation and mapping : ; In the formula, To clean the structure map, As a shoreline structural layer, As a result of water surface suppression, To stabilize the observation confidence level, This is a function for map post-processing or format conversion.

[0016] Compared with existing technologies, this invention has the following advantages: It uniformly describes the reliability of point cloud observations in nearshore waters by using stable observation confidence, avoiding reliance solely on height thresholds or semantic results of single-frame images, thus improving the stability of observation screening in complex nearshore waters; it reduces the probability of surface reflections, wave trails, foam echoes, and false surface points entering the static structure map by utilizing semantic-geometric-temporal consistency, thereby improving the cleanliness of the point cloud map; it enhances the positioning and mapping constraints in weak-structure nearshore environments by utilizing shoreline candidate point extraction and shoreline soft constraints, improving shoreline structure continuity and the stability of UAV pose estimation; it triggers an adaptive robust mapping strategy by jointly evaluating IMU attitude perturbations and environmental observability, reducing the map update weight of low-confidence observations and minimizing abnormal pose jumps and erroneous map expansion in high-wave, attitude-perturbation, and weak-structure regions; and it provides reliable basic data for subsequent UAV navigation map construction, path planning, and obstacle avoidance modules through the outputs of stable structure point cloud maps, shoreline structure layers, water surface instability observation suppression results, and stable observation confidence layers, rather than simply outputting ordinary cumulative point cloud maps without distinguishing observation reliability. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is the overall system block diagram constructed by the present invention; Figure 3 This is a diagram of the stable observation confidence model of the present invention; Figure 4 This is a schematic diagram of soft constraints on the shoreline; Figure 5 This is a schematic diagram of an adaptive robust graph-building state machine; Figure 6 This is a comparison chart of the absolute trajectory errors of various ablation schemes under four nearshore scenarios; Figure 7 This is a comparison chart of the relative pose rotation errors of various ablation schemes under four nearshore scenarios; Figure 8 This is a comparison chart of the relative pose translation errors of various ablation schemes under four nearshore scenarios; Figure 9 These are ablation maps showing the noise ratio of each ablation scheme under four nearshore scenarios; Figure 10 This is a comparison chart of shoreline continuity for various ablation schemes under four nearshore scenarios. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] Methods for stable observation and identification of unmanned surface vessels (USVs) in near-shore navigation mapping and shoreline constraint include: S1. Construct the unmanned vessel state vector and project the laser points of the shipborne lidar onto the image plane; obtain the neighborhood geometric features of the laser points, construct the water surface proximity geometric features based on the neighborhood geometric features; obtain the semantic labels of the laser points; S2. Construct a nearshore observation disturbance index based on disturbance indicators, calculate semantic and geometric consistency scores based on water surface proximity geometric features, semantic labels, and point cloud geometric static structure scores; divide the nearshore space into map units, calculate temporal stability scores based on the hit status of each map unit; construct stable observation confidence scores, and classify observations based on stable observation confidence scores; S3. Based on the confidence level of stable observations, calculate the map update weights, adjust the point cloud matching information matrix, and perform dual suppression on the map update weights; construct an overall objective function that includes inertial measurement unit pre-integration, laser point cloud registration, shoreline soft constraints, and global navigation satellite system residuals for joint optimization; trigger a conservative mapping strategy based on environmental observability, disturbance index, and stable observation ratio, and implement threshold protection for pose translation and heading jump; S4. Construct a clean structure map based on the shoreline structure, construct a shoreline structure layer based on shoreline geometric elements, construct water surface suppression results based on unstable water surface observations, and construct a stable observation confidence layer based on the confidence of each observation unit; by integrating the clean structure map, shoreline structure layer, water surface suppression results, and stable observation confidence layer, the basic data for subsequent navigation mapping is formed.

[0020] S1 includes S1.1, an unmanned surface vessel equipped with a shipborne lidar, inertial measurement unit, camera, and global navigation satellite system; Map coordinate system is The ship's coordinate system is The lidar coordinate system is The camera coordinate system is The coordinate system of the inertial measurement unit is ; Build The state vector of the unmanned vessel at time t is : ; In the formula, This indicates the position of the unmanned vessel in the map coordinate system. For speed, For attitude quaternions, , For accelerometer bias, For gyroscope bias It is the transpose symbol. For discrete time points; Projecting the laser points of the lidar onto the image plane: ; In the formula, Let be the pixel position of the laser point on the image plane. For the camera intrinsic parameter matrix, For perspective projection functions, This is a transformation from the camera coordinate system to the ship's coordinate system. This is a transformation from the lidar coordinate system to the ship's coordinate system. The laser point in the lidar coordinate system. For laser points, Index for laser points.

[0021] S1 includes, S1.2, and the pair. Extracting the neighborhood Calculate the neighborhood mean Covariance Matrix ,right Perform eigenvalue decomposition and define neighborhood geometric features, including linearity. Flatness Dispersion Local roughness High-altitude characteristics and density characteristics Then build Geometric features near the water surface : ; In the formula, It is an exponential function. For absolute values, For high attenuation scale parameters; S1 includes, S1.3, obtaining Semantic tags, including water surface semantic priors Semantic priors of shoreline proximity and static semantic prior .

[0022] S2 includes S2.1, and disturbance indices including the ship attitude disturbance normalization index. Unstable points on the water surface Degree of degradation of weak structures Semantic uncertainty Constructing a nearshore observation disturbance index : ; ; In the formula, , , , These are the weighting coefficients for the four disturbance indicators. This is a truncation function; S2 includes S2.2, which calculates the semantic and geometric consistency score. : ; ; In the formula, , , , , , The weighting coefficients obtained from the calibration, , , , , These are the weighting coefficients in the geometric static structure score. The score is based on the static geometric structure of the point cloud. For the Sigmoid function, For indicator functions, This is a preset height threshold.

[0023] S2 includes S2.3, which divides the nearshore space into map units. For each map unit... Update the static hit status, shoreline hit status, and water surface hit status: ; ; ; In the formula, for Static hit state at all times for Current shoreline hit status. for Constant water surface hit status, As the consistency score threshold, The semantic probability threshold for the shoreline. The linearity threshold, Threshold for water surface geometric features The semantic probability threshold for the water surface. The attenuation coefficient is... For logical AND operator, For logical OR operator; Calculate the timing stability score : ; In the formula, , , This represents the weighting coefficient in the time series stability score.

[0024] S2 includes S2.4, calculating the confidence level of stable observations. : ; In the formula, , , , , , To stabilize the weighting coefficients in the observation confidence score, Scoring for geometric stability, Score for semantic and geometric consistency. The time series stability score is given. The shoreline structure is a priori. The score is for unstable water surface observations. The degree of unreliability caused by attitude disturbances; Set a high confidence threshold and low confidence threshold ; when , The observation was determined to be highly reliable and stable, and entered into the clean structure map to participate in pose optimization. when , The observation is identified as pending confirmation and enters a short-term buffer while awaiting timing confirmation. when , Observations deemed low-confidence will not be included in the static map, or will only be included in the debugging or suppression layer.

[0025] S3 includes, S3.1, based on Calculate map update weights, adjust the point cloud matching information matrix, and apply dual suppression to map update weights: ; ; ; In the formula, for Map update weight, To update the weights for the minimum map size, For maximum map update weight, for Information matrix used in laser point cloud matching To prevent extremely small positive numbers from being zero, for The nominal information matrix, The weights are updated for the final map after double suppression. For point The confidence level for classifying an observation as unstable on the water surface; S3 includes S3.2, constructing the overall objective function: ; ; ; ; In the formula, The sequence of states to be optimized. This is the optimal state sequence. , and To replace the variable, For the pre-integrated residual of the inertial measurement unit, This is the information matrix corresponding to the pre-integrated residual of the inertial measurement unit. For a priori residuals, The information matrix corresponding to the prior residuals, To register residuals for laser point clouds, To register the information matrix corresponding to the residual of the laser point cloud. As the soft constraint weight of the shoreline, For shoreline soft-constraint residuals, For Global Navigation Satellite System residuals, Let be the information matrix corresponding to the residuals of the global navigation satellite system, and be the robust kernel function. For the first The set of indices of high-confidence points in the laser point cloud used for registration at any given time. For the first The set of indices for the shoreline soft-constraint residuals at any given time. For the first A collection of indexes of observations from the global navigation satellite system at any given time. for The index; The optimized pose of the current frame is obtained by solving the overall objective function. .

[0026] S3 includes S3.3, constructing candidate shoreline points. : ; In the formula, To stabilize the observation confidence threshold, The semantic probability threshold for the shoreline. This is the threshold for shoreline hit status. The linearity threshold, For laser points The index of the map unit to which it belongs; right After fitting, a set of shoreline structures is obtained. : ; In the formula, For the first a section of shoreline, This represents the total number of shoreline segments; For the current shoreline point and map shoreline section The distance is: ; In the formula, For the first A section of the map's shoreline. , For shoreline segment indexing, , and for The two endpoints, for Coordinates in map coordinate system M and for and Coordinates in map coordinate system M This refers to the cross product operation of three-dimensional vectors. shoreline soft constraint residuals for: ; In the formula, for arrive distance, This is the normalized scale parameter for the soft constraint distance of the shoreline; shoreline soft constraint weight for: ; In the formula, For shoreline quality, For the current environmental observability, This is the scaling factor.

[0027] S3 includes S3.4, which defines the environmental observability score. : ; In the formula, , , , , , These are the weighting coefficients in the environmental observability score. To normalize the number of edge features, To normalize the number of planar features, To stabilize the observation ratio, For the current frame's shoreline quality, For attitude perturbation. This represents the proportion of unstable points on the water surface. Define conservative mapping strategies, including low-confidence observation weight decay, keyframe insertion interval adjustment, low-confidence area map expansion limit, and unreliable shoreline soft constraint weight reduction; Set environmental observability thresholds Disturbance index threshold and stable observation ratio threshold ,when or or At that time, a conservative mapping strategy was adopted; Define pose translation jump protection: ; In the formula, This represents the change in pose translation between two adjacent frames. To determine the maximum reasonable translation speed for the unmanned vessel, The time interval between two adjacent frames. To protect the tolerance margin during translation; Define pose and heading jump protection: ; In the formula, This represents the change in heading angle between two adjacent frames. The maximum reasonable heading angular velocity of the unmanned vessel. For heading protection tolerance margin; Define pose translation jump protection threshold and pose and heading change protection threshold : ; ; In the formula, To determine the maximum reasonable translation speed for the unmanned vessel, The maximum reasonable heading angular velocity of the unmanned vessel. The time interval between two adjacent frames. To protect the tolerance margin during translation, For heading protection tolerance margin; when or If the conditions are met or ,accept , The update introduces a jump decay factor. And prevent keyframes from being inserted into the current frame: ; when or If the conditions are met or ,reject It then reverts to the pose predicted by the inertial measurement unit.

[0028] S4 includes shore structures such as wharves, quay walls, railings, and embankments; The geometric elements of the shoreline include shoreline points, shoreline segments, shoreline curves, and shoreline framework; Unstable water surface observations include candidate observations of water surface reflection that have been downweighted, delayed in confirmation, or not included in the clean structure map, candidate observations of wave trails, candidate observations of foam echoes, and low-confidence water surface observations. The confidence level of each observation unit includes the observation confidence level of each point cloud point, raster, voxel, or local sub-map unit; Build the foundational data for subsequent navigation and mapping : ; In the formula, To clean the structure map, As a shoreline structural layer, As a result of water surface suppression, To stabilize the observation confidence level, This is a function for map post-processing or format conversion.

[0029] The extrinsic parameters of each sensor in this invention are obtained through offline calibration or online calibration: ; in, This represents the transformation from the lidar coordinate system to the ship's coordinate system. This represents the transformation from the camera coordinate system to the ship's coordinate system. This represents the transformation from the IMU coordinate system to the ship's coordinate system. ; The expression for transforming laser points to map coordinates is: ; In the formula, The coordinates of the laser point in the lidar coordinate system; For each point in the laser point cloud Select its neighborhood Calculate the neighborhood mean Covariance Matrix .

[0030] Neighborhood mean for: ; covariance matrix for: ; right Perform eigenvalue decomposition, assuming the eigenvalues ​​satisfy: ; Define linearity : ; Define flatness : ; Define the dispersion : ; Define local roughness : ; in, To prevent small constants from being divided by zero. A larger value indicates that the neighborhood of the point cloud is close to a linear structure, which may correspond to the shoreline, guardrail, or dock edge. Larger indicates that the neighborhood is close to a planar structure, which may correspond to a quay wall, wharf surface or bank slope; and When the height is large and close to the water surface, it may be due to water surface reflection, wave noise, or instability.

[0031] Definition point height characteristics : ; In the formula, It is a unit vector in the vertical direction. for The average height of the water surface at any given time; Definition point density characteristics : ; High attenuation scale parameter The range of values ​​is Preferred , Its dimensions and height characteristics Consistency is used to control the decay rate of the weights of geometric features near the water surface when the laser point deviates from the water surface height; The standard deviation is not obtained from real-time statistics of the neighboring point cloud.

[0032] This invention utilizes cameras mounted on unmanned surface vessels to acquire forward or panoramic images, and then employs water surface segmentation networks, general semantic segmentation networks, or shoreline region recognition models to obtain semantic probability maps of the water surface region, shoreline adjacent regions, and non-stable water surface structures. This invention is not limited to a specific network; water surface segmentation networks, general semantic segmentation networks, or combinations thereof can be used. For image pixels... Define semantic probability: ; in, Represents pixels The probability of belonging to the water surface area. Represents pixels The probability of belonging to the area adjacent to the shoreline or the waterfront boundary area. Represents pixels The probability of belonging to a non-water surface stable structure region.

[0033] laser point Projected to pixels Subsequently, point-level water surface semantic priors were obtained: ; Obtain point-level shoreline proximity semantic priors: ; Obtain semantic priors for point-level non-water surface stable structures: ; In one alternative implementation, when the semantic model only outputs the water surface probability and shoreline proximity probability, the semantic prior of point-level non-water surface stable structures can be expressed as: ; in, This is an enhancement coefficient for the prior knowledge of stable structures in the area adjacent to the shoreline, used to avoid the shoreline boundary area being misjudged as a low-confidence observation by simple water surface semantics.

[0034] If the laser point If a pixel does not fall within the valid area of ​​the image, or if the semantic confidence of the corresponding pixel is lower than the set threshold, its image semantic prior is set to empty or it is given a default low weight. The stable observation confidence is calculated only based on the point cloud geometric features, multi-frame temporal confirmation results, and attitude perturbation status.

[0035] This represents the observation object to be evaluated, which may include laser points, grid cells, voxel cells, or local submap cells; when x is a laser point... hour, The map unit to which it belongs is denoted as ; , , ; Used to characterize the probability that the observed object belongs to a stable shoreline structure; when For laser points hour, , Calculate using the following formula: ; In the formula, These are the prior weighting coefficients for the shoreline structure. As a semantic prior for shoreline proximity. For linearity, for The shoreline hit status of the corresponding map unit. Geometric features near the water surface; Used to characterize the probability that the observed object belongs to an unstable water surface observation; when For laser points hour, , Calculate using the following formula: ; In the formula, The weighting coefficients for the unstable water surface observation scores. As a semantic prior for the water surface, for Water surface hit status of the corresponding map unit; Used to characterize the impact of attitude perturbations on the reliability of current observations; when For laser points hour, , Calculate using the following formula: ; In the formula, The unreliability weighting coefficient for attitude disturbances. This is a normalized index for ship attitude disturbance. This is the near-shore observation disturbance index. The score is given for the static geometric structure of the point cloud.

[0036] The weighting coefficients of the observed disturbance index satisfy: ; The ship's attitude disturbance parameters can be calculated by the IMU: ; In the formula, , , These are the weighting coefficients. For the first The roll rate at any given moment, For the first The pitch angular velocity at any given moment, For the first Vertical acceleration at time t, It is a neutral acceleration constant; The proportion of unstable points on the water surface can be defined as: ; In the formula, For the first Total number of points in a frame of laser point cloud Threshold for water surface geometric features The semantic probability threshold for the water surface; Semantic uncertainty can be represented by pixel-level probability entropy: ; In the formula, For pixels semantic entropy The pixel position in the image. For a set of semantic categories, for A certain semantic in, Pixels output by the semantic segmentation network Category The probability of; The overall semantic uncertainty of the current image is: ; In the formula, This represents the total number of pixels in the image. when When the value is close to 0, it indicates that the current observation is relatively stable; when When the value is close to 1, it indicates that the current state is one of high perturbation or low observability, and it is necessary to reduce the weight of unreliable observations or enter a conservative mapping mode.

[0037] The point-to-line feature residual of this invention can be expressed as: ; The point-to-surface feature residual can be expressed as: ; in, , Two points on the centerline feature of the map. and These are the map planar parameters.

[0038] Update static hit status, shoreline hit status, and water surface hit status for each map unit. Establish corresponding historical observation state variables, and recursively update these state variables based on the current observation results when each frame of point cloud arrives; the recursive update includes multiplying the state of the previous time step by a decay coefficient. The hit value is incremented based on whether the current frame meets the corresponding judgment condition. In one embodiment, the consistency score threshold for the judgment condition is... The semantic probability threshold for shorelines is The linearity threshold is The water surface geometric feature threshold is The semantic probability threshold for the water surface is .

[0039] when , Determined as a high-confidence, stable observation, it enters the clean structure map and participates in pose optimization, including... Add the index to the first Frame-high reliability stable observation index set : ; Belonging to The weights of the laser points are updated based on the final map. Update clean structure map The corresponding point cloud, raster, or voxel unit; and use it as the laser point cloud registration residual in the overall objective function. The effective observation points. If a laser point simultaneously satisfies the shoreline candidate conditions, it is added to the shoreline soft constraint candidate set for constructing the shoreline soft constraint residual. The normalized scale parameter for the soft constraint distance of the shoreline is: , Its dimension is meters, and it is used to normalize the geometric distance from the current shoreline point to the shoreline segment on the map into a dimensionless residual. The smaller the value, the stronger the constraint effect of the shoreline soft constraint on pose optimization.

[0040] when , Observations identified as pending confirmation are placed in a short-term buffer and await timing confirmation, including... Add to short-term cache set : ; Short-time cache collection is used to store the most recent The index of observation points within the frame that have not yet reached the high-confidence stable observation conditions, along with their corresponding map unit numbers, stable observation confidence levels, and observation times. For observation points awaiting confirmation... Map unit If the following conditions are met within the cache period: ; or: ; Then update the observation point to be confirmed as a high-confidence stable observation point and add its index to the high-confidence stable observation index set. If the number of cached frames exceeds the limit If the above conditions are still not met, the observation point will not be used for cleaning structure map updates, nor will it be used as a laser point cloud registration residual point in the overall objective function.

[0041] when , Observations deemed low-confidence will not be included in the static map, or will only be included in the debugging or suppression layers. Add the index to the low-confidence observation index set : ; Belonging to The laser points do not update the clean structure map. Not included in the high-reliability stable observation index set It is also not used as a laser point cloud registration residual point in the overall objective function; this laser point is only recorded to the water surface suppression result. This is used to characterize suppressed unstable or low-confidence observations of the water surface; In the formula, For low confidence threshold, For high confidence threshold, and ; For short-term buffered frames, This is the static hit confirmation threshold. A shoreline hit confirmation threshold is set. In one embodiment, , , , , .

[0042] In one embodiment, during the process of calculating map update weights, adjusting the point cloud matching information matrix, and applying dual suppression to map update weights, the minimum map update weight is set to... The maximum map update weight is , During the process of constructing candidate shoreline points, a stable observation confidence threshold was set to [value missing]. The semantic probability threshold for shorelines is The shoreline hit threshold is The linearity threshold is .

[0043] Stable observation ratio of the present invention Defined as: ; In the formula, For the first The number of points in the high-reliability stable observation index set. For the first Number of effective laser observation points per frame; Set environmental observability thresholds Disturbance index threshold and stable observation ratio threshold When satisfied or or A conservative mapping strategy was adopted.

[0044] Low-confidence observation weight decay includes, for those that satisfy For low-confidence observations, multiply their final map update weights by a low-confidence decay factor. : ; For simultaneously satisfying For unstable water surface observation points, reset their final map update weight to zero: ; In the formula, For low confidence threshold, For unstable water surface observations, a strong suppression threshold is set. This is the weight decay factor for low-confidence observations.

[0045] Keyframe insertion interval adjustment includes, under normal mapping conditions, adjusting the translation distance between adjacent keyframes when it is greater than... The change in heading is greater than or time interval greater than Insert keyframes as needed; in conservative mapping mode, adjust the keyframe insertion threshold to: ; ; ; In the formula, , and These are the keyframe translation threshold, heading threshold, and time threshold under conservative mapping conditions. , , ; This is the keyframe insertion interval magnification factor. The keyframe is the sequence of states to be optimized that is added. It is also used to update the pose nodes of the local map.

[0046] Map expansion restrictions for low-confidence areas include: no new clean structure map units will be added within low-confidence areas; only existing units that meet the requirements will be allowed. The high-confidence map cells are updated. The low-confidence areas are continuous. Intra-frame average stable observation confidence is lower than or the proportion of stable observations is lower than Map units.

[0047] The weight reduction for unreliable shoreline soft constraints includes multiplying the shoreline soft constraint weight by a shoreline weight reduction factor for unreliable shoreline soft constraints. : ; Alternatively, the shoreline soft constraint can be omitted from the shoreline soft constraint residual set. The above The normalized shoreline quality, determined based on the number of shoreline support points, shoreline segment length, and fitting residual, ranges from 0 to 1. The unreliable shoreline soft constraint is a shoreline soft constraint that satisfies any of the following conditions: shoreline quality. Environmental observability or the number of support points along the shoreline. , , , .

[0048] The inertial measurement unit of this invention predicts pose. The predicted pose is obtained by pre-integration of the inertial measurement unit from the previous stable keyframe to the current frame; the previous stable keyframe is the most recent one that satisfies the stable observation ratio. Environmental observability And the pose change protection keyframe was not triggered.

[0049] This invention defines a pose translation jump protection threshold. and pose and heading change protection threshold : ; ;

[0050] In the formula, To determine the maximum reasonable translation speed for the unmanned vessel, The maximum reasonable heading angular velocity of the unmanned vessel. The time interval between two adjacent frames. To protect the tolerance margin during translation, This represents the heading protection tolerance margin. In one embodiment, , , .

[0051] Optimize pose based on the current frame. Optimize pose with the previous frame Calculate the pose translation change between two adjacent frames. and change in heading angle When satisfied or If a slight pose change is detected in the current frame, the optimized pose for the current frame is accepted. However, the final map update weights for all observation points in the current frame are multiplied by the jump decay factor. And prevent keyframes from being inserted into the current frame: ; In one embodiment, ; When satisfied or If an abnormal pose jump is detected in the current frame, the current frame is rejected from being used to optimize the pose. Update the map and revert to the IMU-predicted pose. .

[0052] The IMU (Inertial Measurement Unit) predicts the pose. The predicted pose is obtained by pre-integration of the inertial measurement unit between the previous stable keyframe and the current frame; the previous stable keyframe is the most recent one that satisfies the stable observation ratio. Environmental observability And the pose change protection keyframe was not triggered.

[0053] The map output by this invention is not a typical cumulative point cloud that does not differentiate between observation reliability and a stable structure map for near-shore positioning and mapping of unmanned vessels. The stable structure map includes: Clean structure map It consists of high-confidence shore structures, wharves, quay walls, railings, embankments, etc. shoreline structural layer It consists of shoreline points, shoreline segments, shoreline curves, shoreline frameworks, or corner points; Results of water surface instability observation suppression For recording or representing water surface reflections, wave trails, foam echoes, and low-confidence water surface observations that have been downweighted, delayed in confirmation, or not included in the static structure map; Stable observation confidence level To record the stable observation confidence of each point cloud point, raster, voxel, or local sub-map unit, used to characterize the observation reliability and the credibility of the map structure; Subsequent navigation mapping basic data It consists of a stable structure point cloud map, a shoreline structure layer, water surface instability observation suppression results, and a stable observation confidence layer. It can be further converted into a raster map, cost map, vector shoreline map, or navigation-aided map according to specific engineering needs.

[0054] The basic data for subsequent navigation mapping can be represented as follows: ; in, This is a map post-processing or format conversion function used to convert stable structure point clouds, shoreline structures, and confidence information into corresponding map representations according to the needs of subsequent unmanned vessel navigation, path planning, or obstacle avoidance modules.

[0055] Unsteady water surface observation refers to the score of unsteady water surface observation. High time series stability score Low or stable observation confidence Lower observation points, grids, voxels, or local submap units.

[0056] When the laser point satisfy and At that time, It was determined to be a candidate observation for water surface reflection; when satisfy , and At that time, It was determined to be a candidate observation for wave trailing wire; when satisfy , and At that time, It was determined to be a candidate observation for foam echo; when satisfy or At that time, The observation was determined to be of low confidence.

[0057] For the above unstable water surface observations, the confidence level of stable observations was used. And water surface instability observation score Perform different processing: Not included in the clean structure map, when or At that time, the clean structure map is not entered; only the water surface suppression results or the debugging layer is entered. Delayed confirmation includes when... When the time is reached, it enters a short-term cache and undergoes a delayed confirmation, pending the temporal stability score of the map unit to which it belongs. Only after meeting the preset temporal stability threshold will they participate in map updates; the reduced weight includes when and At that time, participation in clean structure map updates is allowed, but its map update weight is still based on Perform dual inhibition.

[0058] Water surface inhibition results Used to record unstable water surface observations that have been downweighted, delayed in confirmation, or not included in the clean structure map, and their processing status, including observation location, map unit to which they belong, and unstable water surface observation score. Confidence of stable observations And the corresponding processing markers.

[0059] The relevant parameter settings in this embodiment of the invention include the following: the weighting coefficients for the four disturbance indicators are... , , , The weighting coefficients obtained by preset or calibration are: , , , , , The weighting coefficient in the geometric static structure score is , , , , The weighting coefficients in the time series stability score are: , , The weighting coefficient in the environmental observability score is: , , , , , The above parameters are all standard parameter selections.

[0060] The following description, in conjunction with the accompanying drawings, provides further details. The flowchart of the method of this invention is shown below. Figure 1As shown, LiDAR-IMU observation data is first input, and semantic shoreline projection is completed by combining image semantic results. Then, geometric stability calculation is performed on the local neighborhood of the point cloud to obtain geometric features such as linearity, flatness, and density. Furthermore, semantic shoreline information, geometric stability, and attitude perturbation information are fused to calculate the stable observation confidence, and temporal stability is confirmed by using historical maps or keyframe caches. Afterward, shoreline candidate points are extracted, shoreline soft constraints are constructed, and adaptive weight scaling is performed based on attitude perturbation and environmental observability. Finally, a layered map result is output through robust optimization mapping.

[0061] This invention also constructs an unmanned vessel near-shore navigation, mapping, stability observation, identification, and shoreline constraint system, the overall block diagram of which is shown below. Figure 2 As shown, the sensor input layer acquires multi-source observation data of the unmanned vessel's nearshore waters through lidar point clouds, IMU (Inertial Measurement Unit) motion status, camera shoreline images, and optional GNSS priors, and performs time synchronization, extrinsic parameter calibration, and coordinate unification. Subsequently, the stable observation identification layer combines point cloud geometric stability, semantic shoreline projection, temporal stability confirmation, and attitude disturbance assessment to estimate the confidence level of stable observations. Based on this, the shoreline soft constraint layer further extracts shoreline candidate points, performs shoreline curve or line segment fitting, map shoreline matching, and point-to-shoreline residual calculation to form shoreline soft constraints. The adaptive robust mapping layer completes robust optimization mapping based on environmental observability, keyframe insertion control, adaptive factor weights, and pose jump suppression. Finally, the map output layer outputs a clean structure map, shoreline map, water surface boundary map, confidence map, and localization mapping results.

[0062] The stable observation confidence model of this invention is as follows: Figure 3As shown, this model takes geometrically stable features (shoreline morphology, flatness, density), semantic-geometric consistency, temporal stability (multi-frame hits, continuous occurrence), shoreline structure priors (boundary continuity, distance constraints), unstable water surface observations (water surface probability, far from the shoreline), and attitude perturbation effects (pitch, roll, acceleration) as inputs to perform a stable observation confidence fusion evaluation on the observation reliability of current point cloud points, grids, voxels, or local map units, obtaining a stable observation confidence score. Among them, geometric features are used to determine whether the observation has stable structural features such as linear, planar, or high-density features; semantic-geometric consistency is used to determine whether the image's shoreline partitioning results are consistent with the point cloud's geometric structure; temporal stability is used to determine whether the observation occurs continuously in multiple frames; shoreline structure priors are used to enhance reliable observations with continuous boundaries and close to the shoreline; and unstable water surface observation features and attitude perturbation information are used to reduce the impact of unreliable observations caused by water surface reflection, wave trails, ship roll, and pitch. Based on the confidence level of stable observations, observations are categorized into high-confidence, medium-confidence, and low-confidence observations. High-confidence observations are retained and their mapping weights are enhanced; medium-confidence observations are buffered for short periods and await time-series confirmation; and low-confidence observations have their weights reduced or are suppressed from entering the static map. This confidence level result is further used for shoreline soft constraint construction and adaptive robust mapping control.

[0063] The soft shoreline constraint of this invention is as follows: Figure 4 As shown, shoreline candidate extraction is first performed, which involves extracting shoreline candidate points from high-confidence stable observations and forming the map shoreline structure through clustering, line segment fitting, curve fitting, or skeleton extraction. For the shoreline observation point in the current frame, nearest shoreline matching is performed, i.e., searching for the nearest corresponding shoreline segment in the map shoreline structure; the deviation of the current observation point relative to this shoreline segment is calculated as the shoreline soft constraint residual. This shoreline constraint is not a mandatory constraint, but rather a weighted gating based on the current observation stability, shoreline quality, and environmental observability. When the observation is stable, the shoreline is continuous, and the environmental observability is good, the shoreline constraint weight is high, which can enhance the stability of localization and mapping in weakly structured near-shore scenes; when the shoreline quality is low, the observation is discontinuous, or the environmental observability is poor, the shoreline constraint weight is automatically reduced to avoid erroneous shorelines strongly dragging down pose optimization.

[0064] The adaptive robust graphing state machine of this invention is as follows: Figure 5As shown, the state machine determines whether the unmanned vessel is currently in a stable mapping, disturbance suppression, conservative mapping, or re-stabilization state based on criteria such as attitude disturbance, the proportion of unstable surface observations, the proportion of stable observations, shoreline quality, disturbance index, and environmental observability. In the stable mapping state, the system considers the current environment to have good observability and low disturbance, representing high-confidence observations, and adopts normal map weights, normal keyframe insertion, and normal map update strategies. In the disturbance suppression state, the system detects increased attitude disturbances, increased unstable surface observations, or a rising disturbance index. At this time, it strengthens the robust kernel, limits abnormal pose jumps, and reduces the impact of unreliable observations on the mapping results. In the conservative mapping state, the system considers the current situation to be in a low-observability, weak-structure, or insufficient stable observation scenario. It reduces keyframe insertion, lowers the map update weight of low-confidence observations, and prioritizes the protection of historical shoreline structures to avoid erroneous map expansion. When shoreline observations reappear, shoreline quality improves, and continuous stability is confirmed, the system enters the re-stabilization state, gradually restoring the map update intensity and updating the shoreline model.

[0065] This invention uses lidar point cloud and inertial measurement unit (IMU) data as the main inputs for localization and mapping, and camera semantic results as prior auxiliary information for water areas / shorelines. Through stable observation identification, soft shoreline constraints, and adaptive robust mapping strategies, it completes localization and map construction in nearshore water environments. It was implemented in four scenarios, as shown in Table 1. Table 1 Implementation Scenarios ; All four scenarios mentioned above use the same sensor configuration, the same data processing flow, the same algorithm operation flow, and the same indicator calculation method to ensure comparability between different scenarios.

[0066] The system simultaneously acquires data from LiDAR point clouds, inertial measurement unit (IMU) images, forward-looking camera images, GPS information, and the ground truth pose of the unmanned surface vessel (USV) output by Gazebo. The LiDAR point cloud data is standardized, IMU data is standardized, camera images and intrinsic parameters are synchronized, static extrinsic parameters and TF tree are organized, and time synchronization is monitored. After processing the raw data, a pre-set flight path is established for each scenario, allowing the USV to travel along the pre-set path. The processed data is collected to form a standard data package, and subsequent different results from each scenario are replayed using the same data package. This ensures that differences in results originate only from the activation status of different modules in this invention, and not from flight path, speed, or sensor noise.

[0067] Six control groups, A0 to A5, were set up. All groups used the same ROS bag, the same Gazebo ground truth trajectory, and the same evaluation script, as shown in Table 2. Table 2 Control Group .

[0068] The results are as follows Figure 6 , Figure 7 and Figure 8 As shown, in the V1 rich structure / dock scenario, the ATERMSE (absolute trajectory error) of the A0 baseline method is 3.079m, while the A5 method of this invention reduces it to 0.458m, a reduction of approximately 85.1%. This indicates that in complex environments such as near-shore docks, this invention can reduce the impact of erroneous surface observations and unstable structures on global positioning through stable observation constraints. In the V3 sparse corner weak structure scenario, the ATERMSE of the A0 baseline method reaches 9.868m, indicating that traditional LiDAR-IMU is prone to significant drift in sparse shorelines and weak structure environments. The A5 method of this invention reduces it to 0.546m, a reduction of approximately 94.5%. In the V2 long straight shoreline and V4 open disturbance scenarios, the ATE differences among the methods are relatively small, indicating that the trajectory error index is not sensitive to method differences in these scenarios. However, this invention can still maintain positioning accuracy comparable to the baseline without introducing significant additional drift. The relevant results are shown in Tables 3, 4, and 5. Table 3. Absolute trajectory error data of various ablation schemes in four nearshore scenarios. ; Table 4. Relative pose rotation error data for each ablation scheme in four nearshore scenarios. ; Table 5. Relative pose translation error data for each ablation scheme in four nearshore scenarios. ; In scenario V1, the RPE translation RMSE for A0 is 0.249m, decreasing to 0.122m for A5; the RPE rotation RMSE for A0 is 1.153°, decreasing to 0.397° for A5. This indicates that the present invention not only improves the global trajectory but also enhances the stability of local continuous motion estimation. In scenario V3, the RPE translation RMSE for A0 is 0.297m, decreasing to 0.116m for A5; the RPE rotation RMSE for A0 is 0.668°, decreasing to 0.307° for A5. This demonstrates that in sparse corner and weak structure environments, stable shoreline observation and adaptive robust mapping strategies can reduce the risk of local pose jumps. In V3, the ATE for A4 is 0.852m, and for A5 it is 0.546m; the RPE translation for A4 is 0.370m, and for A5 it is 0.116m; the RPE rotation for A4 is 0.538°, and for A5 it is 0.307°. This demonstrates that the adaptive robust mapping strategy is not merely a decorative module, but rather does indeed further improve localization stability in weakly structured and degenerate scenarios.

[0069] like Figure 9As shown, the method of this invention can significantly reduce the water noise / map noise ratio in all four types of nearshore scenarios, indicating that the proposed stable observation constraint mechanism can effectively reduce the contamination of the final map by water surface reflection, sparse point clouds, and short-term unstable observations, thereby improving the usability of nearshore water point cloud maps. Figure 9 The relevant data is shown in Table 6: Table 6. Map noise ratio ablation results for each ablation scheme in four nearshore scenarios. .

[0070] like Figure 10 As shown, compared to baseline methods and single semantic prior methods, this invention jointly determines stable observations through semantic, geometric, and temporal consistency, enabling the shoreline structure to be continuously maintained during multi-frame mapping, thereby improving the structural integrity of nearshore maps. Figure 10 The relevant data is shown in Table 7: Table 7. Comparison of shoreline continuity results for various ablation schemes in four nearshore scenarios. .

[0071] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for stable observation and identification of unmanned surface vessels (USVs) for near-shore navigation, mapping, and shoreline constraint, characterized in that: include: S1. Construct the unmanned vessel state vector and project the laser points of the shipborne lidar onto the image plane; obtain the neighborhood geometric features of the laser points, construct the water surface proximity geometric features based on the neighborhood geometric features; obtain the semantic labels of the laser points; S2. Construct a nearshore observation disturbance index based on disturbance indicators, and calculate the semantic and geometric consistency score based on the geometric features of the water surface proximity, semantic labels, and point cloud geometric static structure scores. The nearshore space is divided into map units, and the temporal stability score is calculated based on the hit status of each map unit. Construct stable observation confidence scores and classify observations based on these scores; S3. Based on the confidence level of stable observations, calculate the map update weights, adjust the point cloud matching information matrix, and perform dual suppression on the map update weights; construct an overall objective function that includes inertial measurement unit pre-integration, laser point cloud registration, shoreline soft constraints, and global navigation satellite system residuals for joint optimization; trigger a conservative mapping strategy based on environmental observability, disturbance index, and stable observation ratio, and implement threshold protection for pose translation and heading jump; S4. Construct a clean structure map based on the shoreline structure, construct a shoreline structure layer based on shoreline geometric elements, construct water surface suppression results based on unstable water surface observations, and construct a stable observation confidence layer based on the confidence of each observation unit; by integrating the clean structure map, shoreline structure layer, water surface suppression results, and stable observation confidence layer, the basic data for subsequent navigation mapping is formed. S2 includes S2.1, and disturbance indices including the ship attitude disturbance normalization index. Unstable points on the water surface Degree of degradation of weak structures Semantic uncertainty Constructing a nearshore observation disturbance index : ； ； In the formula, , , , These are the weighting coefficients for the four disturbance indicators. This is a truncation function; S2 includes S2.2, which calculates the semantic and geometric consistency score. : ； ； In the formula, , , , , , The weighting coefficients obtained from the calibration, , , , , These are the weighting coefficients in the geometric static structure score. The score is based on the static geometric structure of the point cloud. For the Sigmoid function, For indicator functions, The preset height threshold; S2 includes S2.3, which divides the nearshore space into map units. For each map unit... Update the static hit status, shoreline hit status, and water surface hit status: ； ； ； In the formula, for Static hit state at all times for Current shoreline hit status. for Constant water surface hit status, The consistency score threshold, The semantic probability threshold for the shoreline. The linearity threshold, Threshold for water surface geometric features The semantic probability threshold for the water surface. The attenuation coefficient is... For logical AND operator, For logical OR operator; Calculate the timing stability score : ； In the formula, , , This represents the weighting coefficient in the time series stability score.

2. The method for stable observation and identification of unmanned surface vessels for near-shore navigation mapping and shoreline constraint according to claim 1, characterized in that, S1 includes S1.1, an unmanned surface vessel equipped with a shipborne lidar, inertial measurement unit, camera, and global navigation satellite system; Map coordinate system is The ship's coordinate system is The lidar coordinate system is The camera coordinate system is The coordinate system of the inertial measurement unit is ; Build The state vector of the unmanned ship at time t is : ； In the formula, This indicates the position of the unmanned vessel in the map coordinate system. For speed, For attitude quaternions, , For accelerometer bias, For gyroscope bias It is the transpose symbol. For discrete time points; Projecting the laser points of the lidar onto the image plane: ； In the formula, Let be the pixel position of the laser point on the image plane. For the camera intrinsic parameter matrix, For perspective projection functions, This is a transformation from the camera coordinate system to the ship's coordinate system. This is a transformation from the lidar coordinate system to the ship's coordinate system. The laser point in the lidar coordinate system. For laser points, Index for laser points.

3. The method for stable observation and identification of unmanned surface vessels for near-shore navigation mapping and shoreline constraint according to claim 2, characterized in that, S1 includes, S1.2, and the pair. Extracting the neighborhood Calculate the neighborhood mean Covariance Matrix ,right Perform eigenvalue decomposition and define neighborhood geometric features, including linearity. Flatness Dispersion Local roughness High-altitude characteristics and density characteristics Then build Geometric features near the water surface : ； In the formula, It is an exponential function. For absolute values, For high attenuation scale parameters; S1 includes, S1.3, obtaining Semantic tags, including water surface semantic priors Semantic priors of shoreline proximity and static semantic prior .

4. The method for stable observation and identification of unmanned surface vessels for near-shore navigation mapping and shoreline constraint according to claim 3, characterized in that, S2 includes S2.4, calculating the confidence level of stable observations. : ； In the formula, , , , , , To stabilize the weighting coefficients in the observation confidence score, Scoring for geometric stability, Score for semantic and geometric consistency. The time series stability score is given. The shoreline structure is a priori. The score is for unstable water surface observations. The degree of unreliability caused by attitude disturbances; Set a high confidence threshold and low confidence threshold ; when , The observation was determined to be highly reliable and stable, and entered into the clean structure map to participate in pose optimization. when , The observation is identified as pending confirmation and enters a short-term buffer while awaiting timing confirmation. when , Observations deemed low-confidence will not be included in the static map, or will only be included in the debugging or suppression layer.

5. The method for stable observation and identification of unmanned surface vessels for near-shore navigation mapping and shoreline constraint according to claim 4, characterized in that, S3 includes, S3.1, based on Calculate map update weights, adjust the point cloud matching information matrix, and apply dual suppression to map update weights: ； ； ； In the formula, for Map update weight, To update the weights for the minimum map size, For maximum map update weight, for Information matrix used in laser point cloud matching To prevent extremely small positive numbers from being zero, for The nominal information matrix, The weights are updated for the final map after double suppression. For point The confidence level for classifying an observation as unstable on the water surface; S3 includes S3.2, constructing the overall objective function: ； ； ； ； In the formula, The sequence of states to be optimized. This is the optimal state sequence. , and To replace the variable, For the pre-integrated residual of the inertial measurement unit, This is the information matrix corresponding to the pre-integrated residual of the inertial measurement unit. For a priori residuals, The information matrix corresponding to the prior residuals, To register residuals for laser point clouds, Register the information matrix corresponding to the residual of the laser point cloud. As the soft constraint weight of the shoreline, For shoreline soft-constraint residuals, For Global Navigation Satellite System residuals, Let be the information matrix corresponding to the residuals of the global navigation satellite system, and be the robust kernel function. For the first The set of indices of high-confidence points in the laser point cloud used for registration at any given time. For the first The set of indexes for the soft-constraint residuals of the shoreline at any given time. For the first A collection of indexes of global navigation satellite system observations at all times. for The index; The optimized pose of the current frame is obtained by solving the overall objective function. .

6. The method for stable observation and identification of unmanned surface vessels for near-shore navigation mapping and shoreline constraint according to claim 5, characterized in that, S3 includes S3.3, constructing candidate shoreline points. : ； In the formula, To stabilize the observation confidence threshold, The semantic probability threshold for the shoreline. This is the threshold for shoreline hit status. The linearity threshold, For laser points The index of the map unit to which it belongs; right After fitting, a set of shoreline structures is obtained. : ； In the formula, For the first a section of shoreline, This represents the total number of shoreline segments; For the current shoreline point and map shoreline section The distance is: ； In the formula, For the first A section of the map's shoreline. , For shoreline segment indexing, , and for The two endpoints, for Coordinates in map coordinate system M and for and Coordinates in map coordinate system M This refers to the cross product operation of three-dimensional vectors. shoreline soft constraint residuals for: ； In the formula, for arrive distance, This is the normalized scale parameter for the soft constraint distance of the shoreline; shoreline soft constraint weight for: ； In the formula, For shoreline quality, For the current environmental observability, This is the scaling factor.

7. The method for stable observation and identification of unmanned surface vessels for near-shore navigation mapping and shoreline constraint according to claim 6, characterized in that, S3 includes S3.4, which defines the environmental observability score. : ； In the formula, , , , , , These are the weighting coefficients in the environmental observability score. To normalize the number of edge features, To normalize the number of planar features, To stabilize the observation ratio, For the current frame's shoreline quality, For attitude perturbation. This represents the proportion of unstable points on the water surface. Define conservative mapping strategies, including low-confidence observation weight decay, keyframe insertion interval adjustment, low-confidence area map expansion limit, and unreliable shoreline soft constraint weight reduction; Set environmental observability thresholds Disturbance index threshold and stable observation ratio threshold ,when or or At that time, a conservative mapping strategy was adopted; Define pose translation jump protection: ； In the formula, This represents the change in pose translation between two adjacent frames. To determine the maximum reasonable translation speed for the unmanned vessel, The time interval between two adjacent frames. To protect the tolerance margin during translation; Define pose and heading jump protection: ； In the formula, This represents the change in heading angle between two adjacent frames. The maximum reasonable heading angular velocity of the unmanned vessel. For heading protection tolerance margin; Define pose translation jump protection threshold and pose and heading change protection threshold : ； ； In the formula, To determine the maximum reasonable translation speed for the unmanned vessel, The maximum reasonable heading angular velocity of the unmanned vessel. The time interval between two adjacent frames. To protect the tolerance margin during translation, For heading protection tolerance margin; when or When, if the conditions are met or ,accept , The update introduces a jump decay factor. And prevent keyframes from being inserted into the current frame: ； when or When, if the conditions are met or ,reject It then reverts to the pose predicted by the inertial measurement unit.

8. The method for stable observation and identification of unmanned surface vessels for near-shore navigation mapping and shoreline constraint according to claim 7, characterized in that, S4 includes shore structures such as wharves, quay walls, railings, and embankments; The geometric elements of the shoreline include shoreline points, shoreline segments, shoreline curves, and shoreline framework; Unstable water surface observations include candidate observations of water surface reflection that have been downweighted, delayed in confirmation, or not included in the clean structure map, candidate observations of wave trails, candidate observations of foam echoes, and low-confidence water surface observations. The confidence level of each observation unit includes the observation confidence level of each point cloud point, raster, voxel, or local sub-map unit; Build the foundational data for subsequent navigation and mapping : ； In the formula, To clean the structure map, As a shoreline structural layer, As a result of water surface suppression, To stabilize the observation confidence level, This is a function for map post-processing or format conversion.

Images